every system i've ever monitored did exactly what the code told it to. web, mobile, microservices, gpu clusters.
models are different. it's a live thinking body of neurons, thinking way faster than we can think. & our whole monitoring playbook assumes a human reads the dashboard in time to do something about it.
that assumption is gone. the controls have to sit outside the model & move at its speed, not ours.
i spent years thinking the dashboard was the product.
splunk, then telemetry & reliability for large gpu clusters. in that world, if you could see it & alert on it, you'd done the job.
with models, monitorability is just the means. controllability, safety & alignment are the real benches that get built on top of it. a trace of what the model is doing only matters if something downstream can act on it before the action lands.
most of the conversation right now still stops at "we can see it."
this is a great step into how most of the companies will try to go into production with AI. Monitorability and Observability has truly been essential across each and every workload transition we have had; web2.0, mobile, micorservices, infra, etc. & now we will see a host of companies trying to capitalize across these dimensions as well.
looking forward to what people build next in this space.
nev retweeted
with a lot of push for personal agents, wondering how the economics for consumer businesses will be changing.
all of the existing consumer apps were designed for 1pax (human mostly). with undue agentic pressure, it strains the infra hard, often becoming a heavy line item - ultimately someone will have to bear those margins.
i also see a lot of opportunity for consumer apps to now support transactions or interactions through personal agents, which can also command a bit premium to adjust for increased costs.
crazy example is people using their max possible subscription of Tinder to get maximum matches; on the other side, people also utilizing their personal agents to record maybe netflix?
wdyt?
with a lot of push for personal agents, wondering how the economics for consumer businesses will be changing.
all of the existing consumer apps were designed for 1pax (human mostly). with undue agentic pressure, it strains the infra hard, often becoming a heavy line item - ultimately someone will have to bear those margins.
i also see a lot of opportunity for consumer apps to now support transactions or interactions through personal agents, which can also command a bit premium to adjust for increased costs.
crazy example is people using their max possible subscription of Tinder to get maximum matches; on the other side, people also utilizing their personal agents to record maybe netflix?
wdyt?
era of m&a has begun. congrats to the team!
Listen Labs is joining @Salesforce!
A year and a half ago we launched Listen, an AI platform for understanding customers, and it’s been a crazy ride ever since.
Today, we work with some of the largest companies in the world, including Microsoft, Anthropic, and Sweetgreen.
What started as an AI interviewer is now a full platform for understanding customers. Listen finds the right people, interviews them, analyzes what they say, and even simulates how they'll behave.
Listen’s growth quickly accelerated and while we were raising our next round, we met @Benioff.
Marc is a hero of mine. I've taken countless ideas from Behind the Cloud and watched him framemog every AI CEO on the planet. We're honored to get framemogged next.
With Salesforce, we can bring Listen to every company in the world, much faster.
To our customers, our mission remains the same.I stay CEO, @Florian stays CTO and the same team will keep building the product you rely on, now inside Salesforce AI Labs. Thank you for believing in us early.
To our team, families, and everyone who believed in us before there was much to believe in, thank you.
These years have been the most fun years of my life, and it’s still only just the beginning for Listen.
Now back to work. 🚀
new version of watching hens fight: open-source models playing poker
gave gemma, qwen & llama a very simple instruction:
win every chip. bluff if you want. public table talk. make legal moves.
> qwen immediately started saying way too much in the chat - also revealed his hands
> gemma basically never raised
> llama just kept applying pressure
qwen was chaos. llama was a bully. gemma was just trying to survive.
need to run like 100 of these now.
how does one get unlimited codex rests ? its fun nevertheless
have been trying instinct & @Muse over the past couple of days. pretty good at handling a lot of tasks, but i noticed i wasn't really using them that much.
got me thinking about what i'd actually want ai to handle every day. a lot of it is fairly boring & already sitting on my laptop. also, wanted to handle all of my credit cards - which I was very hesitant to hand-off to @Muse right away. (btw, love that we are only doing one main chat now)
so spent some time seeing what it would take to handle routine things on-device.
> used ollama
> integrated my mail, notes, reminders, calendar
> running locally on my macbook pro
> turns out, pretty impressive
> built using codex (irony, ikr)
helps answer simple questions:
- clean off my emails (yeah, is till have that - lot of spam)
- what did i promise someone i'd get back to?
- what should i know before my next meeting?
- where's that note i wrote about this last week?
- what keeps slipping through the cracks?
individually, pretty small things. collectively, a lot of opening apps, finding context & remembering what you were doing.
going to keep using it for a bit. – at Palo Alto, CA
thought of sharing some nuggets from getting washed away in crazy deployments. i bet people are running into these challenges even at frontier.
